FormalPara Key Points

The study presents first evidence on the penetration of bevacizumab into the cerebrospinal fluid (CSF), with a CSF/plasma ratio of 0.16.

Malignant cells were present in CSF before (5/12 patients), but not after treatment.

Vascular endothelial growth factor A (VEGF-A) was detected in three patients before treatment and in one out of these three patients despite treatment.

1 Introduction

Brain tumors represent the most common form of solid malignancies in the pediatric population and substantially contribute to cancer-related morbidity and mortality in children and adolescents [1]. Despite multimodal treatment strategies including surgery, radiotherapy, and chemotherapy, approximately 25% of pediatric brain tumor patients experience relapse or progression, often leading to fatal outcomes. The histological categories most prone to progression or recurrence comprise high-grade glioma, ependymoma, medulloblastoma, and atypical teratoid rhabdoid tumor (ATRT) [2].

Treating recurrent brain tumors is complicated by their tendency to leptomeningeal dissemination. Drug concentrations within the central nervous system (CNS) are influenced by multiple factors, including the permeability of the blood–brain barrier (BBB) and the blood–cerebrospinal fluid (CSF) barrier to the therapeutic agent. Despite the critical role of CNS distribution in the treatment of brain tumors, scarce information exists for many administered drugs, particularly monoclonal antibodies, regarding their concentration–time profiles in the CSF. The penetration of antibodies into the CSF has been reported to be approximately 0.1% of the serum concentration in patients [3,4,5]. Given the wide use of antibodies as antineoplastic agents, understanding their distribution in the CNS and CSF is imperative for the effective use of antibody-based therapeutics against brain tumors [6].

Malignant cells often exhibit a propensity to develop drug resistance, posing a major challenge in managing recurrent malignancies. Beyond cancer cells themselves, the tumor microenvironment plays a critical role in tumor progression and response to treatment across various cancer types. Metronomic antiangiogenic therapy exerts its antitumor effects through multiple mechanisms, by targeting the tumor microenvironment via the inhibition of angiogenesis, by modulating tumor immunity, and by directly targeting cancer cells [7,8,9,10,11,12]. The development of metronomic antiangiogenic therapy as a treatment option emerged from the observation that tumors resistant to dose-intense therapy responded to the same agents at low continuous dosing [7, 8]. Metronomic antiangiogenic therapy regimens including bevacizumab are increasingly applied in the treatment of various pediatric CNS malignancies, particularly in recurrent tumors, and have shown encouraging treatment effects [13, 14]. Bevacizumab is a recombinant humanized monoclonal IgG1 antibody that binds to and inactivates vascular endothelial growth factor A (VEGF-A), thereby inhibiting angiogenesis [15]. A multicenter, controlled trial recently showed prolonged survival in patients with recurrent medulloblastoma using a metronomic antiangiogenic regimen supplemented by intravenous bevacizumab [16].

Children and adolescents and young adults (AYA) frequently exhibit different drug exposure and may require varied drug dosages compared to adults to achieve the desired therapeutic effect. Despite the evident need for investigated safe treatment regimens for children and adolescents with malignant diseases, most drugs have exclusively been approved for adult patients. Furthermore, AYA aged between 15 and 39 years represent a frequently overlooked group [17,18,19].

Ethical considerations and practical challenges frequently constrain the inclusion of children and AYA in clinical trials [20], resulting in a lack of comprehensive data regarding the pharmacological behavior of drugs within these age groups. Furthermore, sampling is limited in clinical studies involving children, leading to a scarcity of data that poses challenges for data analysis. Population pharmacokinetic models have proven particularly useful in situations with sparse and/or unbalanced data (i.e., with few observations per patient sampled at different times of the dosing interval), as often encountered in pediatric clinical trials [21]. These models do not only allow determination of pharmacokinetic parameters for a population and individual patients, but also quantification of variability in the pharmacokinetic parameters and concentration–time profiles between patients.

Several studies have investigated the pharmacokinetics of bevacizumab in children in serum [22, 23]. In pediatric cancer patients, bevacizumab exhibited linear pharmacokinetics and a half-life of 19.6 days [23]. However, little is known about the pharmacokinetics and activity of bevacizumab in the CNS. Because an intact BBB restricts the penetration of monoclonal antibodies into the CNS, it has been assumed that bevacizumab might be unable to enter the CSF.

The present study aimed to address this question and quantified bevacizumab concentrations in serum and CSF in children and adolescents with recurrent malignant brain tumors. A population pharmacokinetic model was employed to assess the extent of distribution into the CSF. We further investigated the impact of bevacizumab on VEGF-A levels in CSF and the presence of cytological leptomeningeal metastases.

2 Patients and Methods

2.1 Patients and Treatment

Pediatric patients with a histologically confirmed diagnosis of a malignant brain tumor receiving metronomic antiangiogenic therapy were eligible for the study. All enrolled patients had an Ommaya reservoir in place. Bevacizumab (Avastin®, 500 mg, concentrate for preparation of an infusion solution, Roche, Switzerland) was administered as intravenous infusion at a dosage of 10 mg/kg over 30 min every 2 weeks. Patients further received thalidomide (3 mg/kg/day daily, orally), celecoxib (50–400 mg twice daily, orally), fenofibrate (90 mg/m2 daily, orally), alternating 21-day courses of etoposide (35–50 mg/m2 daily, orally), and cyclophosphamide (2.5 mg/kg daily, orally). Additionally, intraventricular etoposide was administered (0.5 mg × 5 days in 1 week out of every 4 weeks), alternating with intraventricular liposomal cytarabine (25–50 mg every 4 weeks) [13, 14, 16].

Further inclusion criteria encompassed (1) a life expectancy of at least 8 weeks, (2) the acquisition of written informed consent from the parents and/or the patients themselves, (3) serum creatinine concentrations < 1.5 mg/dL, (4) total serum bilirubin concentrations < 2.0 mg/dL and alanine aminotransferase concentrations < 5 times upper limit of normal, and (5) platelet counts > 40,000/mm3 within 48 h preceding the first treatment. The study was approved by the Ethics Committee of the Medical University of Vienna, carried out in accordance with the Declaration of Helsinki, and registered in the European Clinical Trials Database (EudraCT number 2009-013024-23).

2.2 Sampling, Bioanalysis, and Cytology

For the pharmacokinetic study, blood and CSF samples were collected under real-world conditions defined by the patient’s treatment protocol. Approximately 1.2 mL of venous blood were collected in tubes without previously added anticoagulant before and within 3 h following bevacizumab administration. CSF (2 mL) was collected in separate tubes after puncture of the Ommaya reservoir at steady state, immediately before the intraventricular administration of antineoplastic drugs (etoposide or liposomal cytarabine). To avoid any additional risks for the patients, CSF samples were only taken on occasions when the Ommaya reservoir was punctured for routine reasons. Blood samples were kept on ice for a maximum duration of 60 min and centrifuged at +4 °C and 3000 rpm for 10 min. CSF and serum samples were snap frozen and stored at −80 °C until analysis.

Bevacizumab concentrations were determined using an enzyme-linked immunosorbent assay (ELISA), according to the manufacturer’s instructions. Briefly, snap frozen serum and CSF aliquots were thawed on ice before analysis in triplicates. The lowest quantifiable concentration in human serum and CSF was 30 ng/mL (Matriks Biotech, Ankara, Turkey), and detected values were validated with a calibration curve.

CSF concentrations of human VEGF-A taken before the start of recurrence treatment and the last available sample at the end of treatment (eight patients at the end of planned treatment, four patients due to progression) were determined using a specific sandwich ELISA kit (RayBiotech, GA, USA; catalog number ELH-VEGF) following the standard procedure specified by the manufacturer. The lowest quantifiable concentration in human serum and CSF was 10 pg/mL. Samples were used without dilution, and the absorption was measured at 450 nm.

A cytological examination of CSF was performed using cytospins stained with a May-Grünwald-Giemsa staining. The results were reviewed by an experienced neuropathologist to identify any presence of malignant cells.

2.3 Pharmacokinetic Assessment of Serum and CSF Concentrations

Population pharmacokinetic modelling was used to describe CSF concentrations of bevacizumab relative to serum concentrations. The population approach has demonstrated substantial utility in analyzing scenarios with sparse and unevenly distributed data. A robust and externally validated population model for pediatric cancer patients was used to describe bevacizumab concentration–time profiles in serum [23]. The two-compartment model was developed based on a large patient population (152 patients, 1427 observations) spanning broad ranges of age (0.5–21 years) and body weight (5.9–125 kg). The patients underlying the model received a similar dosing regimen (5–15 mg/kg biweekly or triweekly) as the current study population.

To obtain predictions of serum concentrations for the study participants, parameters from children with primary CNS tumors (displaying similar characteristics as the enrolled patients) were used and fixed to their final estimates (Table 1). The model contained covariate effects (i.e., patient characteristics influencing the pharmacokinetics of bevacizumab) of albumin, sex, and tumor type on clearance, tumor type on central volume of distribution, and body weight on all distribution and elimination parameters (using allometric or power functions). Individual pharmacokinetic parameters and, consequently, serum concentration–time profiles were obtained post hoc for each patient, based on their individual covariate values and the available two serum concentrations.

Table 1 Parameter estimates of the population model
Full size table

To assess drug penetration into the CSF, the model was adapted to accommodate CSF concentrations of bevacizumab by adding an additional peripheral compartment to the serum-based pharmacokinetic model (Fig. 1). Consequently, the parameters volume of distribution of CSF (VCSF), intercompartmental clearance between the central and CSF compartment (QCSF), and an uptake factor, accounting for bevacizumab penetration from the central compartment into the CSF (‘CSF/serum ratio’), were introduced [26].

Fig. 1
figure 1

Schematic depiction of the population pharmacokinetic model used to characterize serum and CSF concentrations of bevacizumab. CL clearance; CSF cerebrospinal fluid; Q/QCSF intercompartmental clearance parameters; UPTK uptake factor capturing drug penetration into CSF; VC volume of distribution of the central compartment; VP volume of distribution of the peripheral compartment; VCSF volume of distribution of the CSF compartment

Full size image

Between-patient variability of the model parameters was described assuming a log-normal distribution. Proportional, additive, and combined error models were explored to describe residual variability. For model discrimination, a decrease in the objective function value (OFV) of ≥ 3.84 was deemed significant (given 1 degree of freedom and α = 0.05 according to the χ2 distribution). Model evaluation included standard goodness-of-fit plots (e.g., observed vs [individual] predicted concentrations, conditional weighted residuals vs predictions and time), as well as a visual predictive check (n = 1000) to assess the predictive performance of the model. A non-parametric bootstrap (n = 500) was performed on the final model to assess parameter uncertainty.

Stochastic simulations (n = 500) were performed with the final model to assess bevacizumab penetration into CSF considering interindividual variability in the pharmacokinetic parameter estimates. Areas under the concentration–time curve (AUCs) were compared for a dosing interval of 2 weeks after 20 doses of 10 mg/kg for 12 individuals with patient characteristics of the study participants.

All modelling activities were conducted using NONMEM®7.4 (ICON Development Solutions, Hanover, MD, USA; first-order conditional estimation method with interaction [FOCEI]) and supported by PsN (version 4.9.6, Uppsala University, Uppsala, Sweden). Model evaluation and graphical illustrations were carried out using R (v3.6.2., R Foundation for Statistical Computing, Vienna, Austria; https://www.R-project.org).

3 Results

3.1 Patients

The pharmacokinetic study was performed in 12 patients (five female, seven male) with a median age of 13 years (range 5–27 years; see Table 2 for further patient characteristics). Diagnoses were medulloblastoma (n = 5), ependymoma (n = 1), choroid plexus carcinoma (n = 1), pineoblastoma (n = 2), CNS germ cell tumor (n = 1), CNS embryonal tumor, not otherwise specified (n = 1), and ATRT (n = 1). All patients received treatment for recurrent disease.

Table 2 Patient characteristics of the study population
Full size table

3.2 Pharmacokinetic Assessment of Serum and CSF Concentrations

Serum and CSF concentrations were obtained following 22 (median) (min–max: n = 7–41) prior doses of bevacizumab. Bevacizumab administration (10 mg/kg) and measurements in serum were performed 14 days (range 12–15 days) after the previous infusion of bevacizumab. Trough (minimum concentration [Cmin]) and peak (maximum concentration [Cmax]) serum samples (before and after intravenous infusion of bevacizumab) were available for all patients (nsample = 24). Cmin ranged from 77.0 to 305 mg/L (median 144 mg/L), and Cmax encompassed values between 267 and 612 mg/L (median 417 mg/L; Fig. 2).

Fig. 2.
figure 2

Left panel: Serum concentrations, measured before and after bevacizumab (BVZ) infusion (vertical red line). Right panel: concentrations in cerebrospinal fluid (CSF), measured on different days following BVZ administration (partly in different dosing intervals). BLQ below the lower limit of quantification

Full size image

CSF samples were taken when intraventricular therapy was scheduled. In each patient, three (median) (min–max: 3–4) CSF concentrations were obtained during different treatment cycles of bevacizumab within 1–15 days after dosing (see Fig. S1 in the Online Resource [electronic supplementary material] for individual CSF and serum sampling). In CSF, concentrations of 0.01–2.26 mg/L (median 0.35 mg/L) were quantified (ntotal = 38; 3–4 samples/patient; Fig. 2).

Individual pharmacokinetic parameters were determined for each patient using a previously developed two-compartment model for bevacizumab [23]. The model predicted Cmin and Cmax concentrations reasonably well (Fig. 3; individual profiles: Fig. S2, Online Resource) and was used to describe the serum concentration–time profiles of the patients also at later time points (Fig. 4).

Fig. 3
figure 3

Goodness-of-fit plots showing the ability of the model to predict measured Cmin and Cmax of bevacizumab in serum based on patient characteristics (‘covariates’). The solid black line represents the line of identity. Cmax maximum concentration, Cmin minimum concentration

Full size image
Fig. 4
figure 4

Measured serum concentrations (blue circles) and predicted serum concentration–time profiles of bevacizumab (BVZ) based on a previously published population pharmacokinetic model for pediatric cancer patients [23]

Full size image

To quantify bevacizumab penetration into CSF, an additional compartment was added to the model (Fig. 1). The difference between serum and CSF concentrations was modelled using an uptake factor (UPTK), which was 0.16 for the typical patient in the population. The model significantly improved when allowing interindividual variability on the uptake factor (ΔOFV 48.5) and revealed a broad range of individual estimates between 0.02 and 0.66 (median 0.17). Simulations considering between-patient variability in the pharmacokinetic parameters revealed a median ratio of AUCPL/AUCCSF of 0.156 (95% prediction interval 0.00282–0.897).

Due to the sparse nature of the data, two CSF parameters were fixed to physiology-based values. VCSF was set to 150 mL/70 kg with linear weight scaling [27]. Intercompartmental clearance to CSF (QCSF) was described as the blood flow rate to brain (12% of cardiac output, corrected by hematocrit to represent plasma flow) [24, 28]. Furthermore, a blood flow correction factor (FCSF) was added to the model, but did not lead to significant model improvement. FCSF was estimated as 0.33, indicating permeation restriction. Residual unexplained variability was described using a proportional model, which was superior to an additive model (ΔOFV 37.3) and not inferior to a combined variability model. Model parameter estimates are summarized in Table 1. To assess a potential influence of one CSF concentration below the lower limit of quantification on the results, the model was rerun after exclusion of this specific sample. Model parameters remained virtually unchanged (UPTK 0.0016; relative standard error 27.2→25.2%), with just a slight reduction of interindividual variability associated with UPTK, as expected (coefficient of variation [CV%] 107→101%).

Goodness-of-fit plots indicated good agreement between the observed and predicted data (Fig. 5), without any striking trends and with random scatter around the line of identity or reference line (conditional weighted residuals: Fig. S3, Online Resource). Individual predicted profiles (Fig. S4, Online Resource) were reasonably predicted given the sparse data situation, and the visual predictive check showed acceptable predictive performance of the model, with slight overestimation of interindividual variability (Fig. 6).

Fig. 5
figure 5

Goodness-of-fit plots showing the ability of the model to predict measured concentrations (conc.) in the cerebrospinal fluid (CSF). The solid black line represents the line of identity

Full size image
Fig. 6
figure 6

Visual predictive check of the model for cerebrospinal fluid concentrations of bevacizumab (BVZ). Blue circles represent measured BVZ concentrations; the blue solid line shows median observed concentrations, and the blue dashed lines indicate the 10th and 90th percentiles of the observed data. The green solid line shows the median of model predictions. Shaded areas depict the 95% confidence intervals for the prediction lines (median, 10th and 90th percentile)

Full size image

3.3 VEGF-A Concentrations in the CSF

VEGF-A was detected before the start of treatment for the recurrent tumor in three patients (25%), diagnosed with medulloblastoma, plexus carcinoma, and pineoblastoma. VEGF-A concentrations ranged from 12 to 33 pg/mL (mean ± SD: 20 ± 11 pg/mL). After the treatment, VEGF-A levels were below the detection limit in all but one patient (Table 3).

Table 3 VEGF-A and cytology at recurrence and last available examination during the course of disease
Full size table

3.4 Cytology

Upon recurrence, before initiation of treatment, malignant cells in the CSF were detectable in five out of 12 patients. At the time of pharmacokinetic sampling as well as at the last available examination at the end of treatment, all patients attained treatment response, with CSF samples tested negative for malignant cells. Clearance of VEGF-A levels and clearance of malignant cells in CSF cytology correlated in two cases (Table 3). Overall, detection of tumor cells in CSF and detectable VEGF-A levels were not significantly associated (Fisher’s exact test, p = 0.5).

4 Discussion

This study presents a comprehensive analysis of bevacizumab in serum and CSF in children and AYA with recurrent brain tumors using population pharmacokinetic modeling and simulation. Metronomic antiangiogenic therapy including bevacizumab is increasingly used in recurrent brain tumors in children and AYA [13, 14].

The BBB prevents the entry of many therapeutic agents to the CNS. The tight junctions interconnecting the endothelial cells of the BBB restrict the free passage of water-soluble molecules and cells [29], prevent the leakage of substances from the bloodstream into the CNS, and thereby protect the brain against circulating potentially neurotoxic substances. The BBB serves multiple functions in neuroimmunology by acting as a barrier, isolating the CNS from the immune components of the periphery. This unique function grants the brain its status as an immunologically privileged organ [30]. Although the intact BBB is expected to restrict the penetration of molecules with high molecular weight into the CNS, there is some evidence suggesting that the permeability of the BBB may be increased in patients with impaired BBB, e.g., those having undergone cranial radiotherapy [31].

To date, no direct assessment of the extent to which intravenous bevacizumab can reach the CSF has been performed. To address this important clinical question, we measured bevacizumab in serum and CSF and assessed the extent of distribution using population pharmacokinetic modelling and simulation. Serum concentration measurements were available only shortly before and after bevacizumab administration (Cmin 77.0–305 mg/L; Cmax 267–612 mg/L). The observed serum concentrations were in the range of previous studies in children with osteosarcoma [32] and patients with advanced non‑squamous non‑small cell lung cancer [33]. These studies showed variable serum concentrations up to 800 mg/L (even in the first cycles) given 15 mg/kg dosing over 30–90 min, corresponding to 530 mg/L for 10 mg/L assuming dose linearity, as had been shown for bevacizumab within a dose range of 1–20 mg/kg [34]. In a study with dosing ≤ 20 mg/kg infused over 30–90 min (600 mg/L assuming 10 mg/kg dosing), serum concentrations even up to 1200 mg/L have been reported [34]. A previous study with eight children with refractory solid tumors demonstrated lower Cmax of bevacizumab than in the present investigation (230–425 mg/L after 15 mg/kg, corresponding to 153–283 mg/L after 10 mg/kg) [35]. However, like in the aforementioned studies, longer infusion durations (30–90 min vs 30 min), leading to lower Cmax, and sampling in early phases rather than at steady state could explain this difference. None of the studies investigated CSF concentrations.

The current study found measurable bevacizumab concentrations of 0.01–2.26 mg/L (median 0.35 mg/L) in CSF. Unlike serum concentrations, CSF samples were not exclusively taken at the exact time points of the serum concentrations, which covered merely a small segment of the concentration–time profile. Thus, serum pharmacokinetics of the study participants were predicted using a previously published two-compartment model of a pediatric and adolescent population with primary CNS tumors, and CNS concentrations were described based on individual predicted serum profiles. An uptake factor of 0.16 was estimated as a measure of the fraction of bevacizumab concentrations penetrating from the central compartment into the CSF. This magnitude was in agreement with penetration ratios derived from raw data of serum and CSF concentrations measured on the same day. The estimated uptake factor aligns well with physiological CSF/serum ratios (0.16–0.32 [36, 37]) and the limited uptake of therapeutic antibodies (0.1–0.2% of injected dose) into the brain [38]. CSF penetration has been demonstrated for several antibodies, including ponezumab (0.5% to < 1% [39, 40]), elezanumab (0.1–0.4% [41]), nivolumab (0.3–1.9% [42, 43]), atibuclimab (0.03% [44]), GNbAC (0.12–0.42 [45]), and rituximab (0.1–0.3%) [4]. For bevacizumab, previous evidence on penetration into the CNS is scarce. A study with mice showed bevacizumab distribution within brain structures and particularly within glioblastoma tumor areas after intraperitoneal injection [46]. However, the transferability of these results to humans is not clear, and although also a surrogate for CNS penetration, distribution to brain or tumor tissue cannot be directly compared to CSF penetration.

The measured CSF concentrations (0.01–2.26 mg/L) partly exceeded the inhibitory concentrations reported in several previous experimental studies, suggesting a potential for effects in this peripheral compartment. For example, investigations of the binding affinity of bevacizumab for VEGF isoforms demonstrated concentration for 50% of maximal effect (EC50) values ranging from 0.15 nM (0.0224 mg/L given a molar mass of 149,196.8162) [47] to 0.18 mg/L [48]. Fifty percent inhibitory concentration (IC50) values for inhibiting vascular endothelial growth factor receptor 1 (VEGFR1) and VEGFR2 binding/activation lay between 0.63 nM (~ 0.0940 mg/L), 1.476 nM (~ 0.220 mg/L), and 2.661 nM (0.397 mg/L) [15, 49]. Furthermore, bevacizumab inhibited the proliferation of bovine retinal microvascular endothelial cells with an average IC50 of 0.668 nM (~ 0.0997 mg/L) [50]. Similarly, assays with human umbilical vein endothelial cells (HUVEC) showed that VEGF-induced proliferation was blocked with IC50 = 0.047 ± 0.0081 mg/L [48], and VEGF-A165-induced calcium mobilization was inhibited with IC50 = 70.8 pM (0.0106 mg/L) [15], thus inhibitory concentrations were within the magnitude of the measured CSF concentrations.

The detection of tumor cells in the CSF remains the gold standard for a definite diagnosis of leptomeningeal dissemination. However, the cytological analysis of the CSF suffers from several important shortcomings [51]. These limitations are especially relevant in children, in whom only small CSF volumes can be collected.

The secretion of the proangiogenic molecule VEGF-A has been recognized as a crucial mechanism through which cancer cells stimulate neoangiogenesis in tumor tissue [15]. Patients with malignant tumors and leptomeningeal dissemination showed high values of VEGF in CSF [52, 53]. During treatment with radiation and intrathecal chemotherapy in adult patients with carcinomatous meningitis, responders showed a decrease in elevated CSF VEGF concentrations below the detection limit, while non-responders maintained high concentrations [52]. In our study’s patients with recurrent brain tumors, VEGF-A could be detected in the CSF of three patients, all of whom also exhibited positive CSF cytology, reflecting the metastatic nature of their disease. While the CSF could be cleared of malignant cells during treatment in all patients, detectable VEGF-A persisted in one patient. This might be a further indication of the limitation of CSF cytology in diagnosing leptomeningeal disease, even with repeated sampling.

Several limitations should be acknowledged in the context of this study. First, the number of patients in the study was rather small (n = 12). Second, given the limited number of blood draws to reduce the burden on children, the available data were sparse, with merely two serum concentrations (both collected right at the initiation of the dosing interval) and two to three CSF samples per patient. A previously published model built upon rich data (n = 1427 observations) was thus used to predict complete serum concentration–time profiles as the basis for describing CSF concentrations relative to serum concentrations [23]. The assumption that serum pharmacokinetics follow the model described by Han et al. [23] was deemed reasonable, as the population underlying the model was largely comparable with the study population, with both pediatric and adolescent patients with primary CNS tumors, similar administered dosages (5–15 mg/kg biweekly or triweekly ↔ 10 mg/kg), and virtually identical median body weight (44 ↔ 47 kg) and albumin concentrations (39 ↔ 39 g/L). Two patients displayed higher ages (26 and 27/0.5–21 years). However, pharmacokinetic differences in this narrow age range of young adults are likely small and model evaluation did not reveal worse performance for these individuals. As a further limitation, the CSF data obtained in this study were not sufficiently rich to support the estimation of the volume of distribution of CSF and the intercompartmental clearance between CSF and the central compartment. Thus, VCSF and QCSF were fixed to physiology-based values in line with multiple previous studies on CSF pharmacokinetics [26, 54]. Studies with more rich data and more patients would be useful to describe CSF penetration of bevacizumab in more detail and with higher precision, as well as to assess a potential impact on tumor type on CSF penetration.

5 Conclusions

In conclusion, this pharmacokinetic pilot study indicated bevacizumab penetration into the CSF in a population of children and AYA with recurrent brain tumors. Further studies involving more patients and more rich data will be interesting to confirm results and to investigate factors influencing drug distribution to the CSF in children and adolescents with recurrent brain tumors.